OntoRAG: Automating Ontology Derivation for Enhanced Question-Answering

From Unstructured Electrical Relay Documents to Structured Ontologies

Published

May 31, 2025

Authors: Y. Tiwari et al.
Published on Arxiv: 2025-05-31
Link: http://arxiv.org/abs/2506.00664v1
Institutions: ABB Ability Innovation Center, Bangalore, India • ABB Ability Innovation Center, Hyderabad, India • Indian Institute of Technology, Kharagpur, India
Keywords: LLM, Ontology Learning, Retrieval-Augmented Generation, Knowledge Graph, GraphRAG, Leiden algorithm, Semantic Web, Question Answering, PDF Parsing, Information Extraction

Random Unsplash-style image

Ontologies are crucial for structuring knowledge bases to improve the performance of question-answering (QA) systems using Large Language Models. However, the manual creation of ontologies is a significant bottleneck, especially in large or rapidly evolving technical domains.

To address these challenges, the authors introduce their solution and methodology as follows:

Following the description of OntoRAG’s methodology, the results further underscore its efficacy:

To complete the discussion, the authors offer their main conclusions: